The bigger picture of rawspeed: raw data versus metadata?

I know that RawSpeed intentionally does not extract metadata (like thumbnails, EXIF data, and alike) out of the raw files. LibRaw on the other hand does this. Now I'm wondering how one should go about using RawSpeed if you want the fast decoding performance of RawSpeed, but still want to extract a thumbnail for example; and have access to all the other metadata?

It feels bloated to run both LibRaw and RawSpeed on the same file. Reading the file twice, and parsing the file twice. DarkTable probably can extract the metadata. How is it done; and is it considered fast?

I was hoping that instead of providing us with the metadata itself, RawSpeed could tell us the byte-offsets in the file of several metadata sections. For example: the JPEG-thumbnail sits at offset 518 and is 420230 bytes in length. Same for a potential EXIF section and so forth. If this strategy makes sense at least.

So in general: How does one use RawSpeed and get the metadata (like photographer name, time the photo was taken, GPS coordinates, etc etc)? I think the answer to this question should probably be somewhere in the README?

Start with the README and the issue's comparison of RawSpeed, LibRaw, and darktable. Clarify and document the supported way to obtain metadata and thumbnails alongside RawSpeed decoding, including whether the proposed metadata-section offsets are part of the intended scope; done means the README answers the usage question.